What is mattwood.fyi?

A live, structured feed of riffs and links from Matt Wood, Chief AI & Technology Officer at AWS — readable by people and queryable by agents as a typed knowledge graph.

Is there an API and is it free?

Yes. A public, read-only query API is available with no authentication and no API keys. See https://mattwood.fyi/openapi.json and https://mattwood.fyi/docs.

How should an AI agent use this site?

Fetch https://mattwood.fyi/feed.json for the full structured feed, or query the API for keyword/semantic search and the connection graph. See https://mattwood.fyi/agents/ for when-to-use guidance.

Can LLMs Play NetHack? A 2026 Agent Benchmark

Explores whether modern large language models can play NetHack, comparing LLM-based agents against symbolic and neural bot approaches, and presents a custom agent harness with experimental results.

kenforthewin.github.io →

Connections

Related to

Dream-RSI: Recursive Self-Improvement through Evolving Worlds Dream-RSI's recursive self-improvement through evolving worlds shares conceptual overlap with NetHack as a benchmark environment — both treat complex, partially-observable worlds as testbeds for agent learning and improvement
An LLM Beat NetHack Both directly concern LLM performance on NetHack; the benchmark paper provides systematic evaluation while the other reports a specific achievement, making them complementary assessments of the same problem domain
Unreal Agent: Cost-Efficient AI Agent Harness Both involve custom agent harness design for LLM evaluation; the NetHack paper presents a custom agent harness, and Unreal Agent focuses on cost-efficient agent harness construction

Supports

SWE-bench Science: Coding Agents in Scientific Software Engineering Both papers present structured benchmarks for LLM-based agents in challenging domains (game-playing vs. scientific software), contributing to the broader empirical literature on agent capabilities and limitations
Real-SWE Benchmark Both are rigorous LLM agent benchmarks comparing against baseline approaches; the NetHack benchmark complements Real-SWE by extending agent evaluation to complex, open-ended game environments rather than software engineering
Why ML Research Agents Don't Overfit The NetHack benchmark empirically tests whether LLM agents generalize in a complex environment, directly relevant to the question of why ML research agents avoid overfitting in open-ended evaluation settings

Type:	link
First published:	Friday, 25 September 2026 at 22:50 UTC
Location:	https://kenforthewin.github.io/blog/posts/nethack-agent/
URL:	https://mattwood.fyi/i/d97531f7

I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.

Connections